I agree that getting an exact 100% average over multiple minutes is going to be 
somewhat unlikely. But we can get close.

From RMF's Partition Data Report:

--- PHYSICAL PROCESSORS ---
LPAR MGMT  EFFECTIVE  TOTAL
    0.02       8.60   8.62 
    0.12      78.81  78.93 
    0.05      11.53  11.58 
    0.01       0.50   0.51 
    0.12              0.30 
   -----      -----  ----- 
    0.31      99.44  99.93 

That's one 15 minute interval, but 5 consecutive intervals were above 99%, only 
decreasing when the group cap hit and dropped the utilization down to about 76% 
of the physical machine. It ran at the group cap for around 6 hours after that, 
which you can kind of consider running at 100% of the "available" capacity. 
(Actually more than that, since the group cap is set to about 74% of the 
machine; the caps aren't enforced perfectly.)

Taking away about a quarter of the capacity via a group cap may seem a bit 
extreme, but the batch window finished with hours to spare, and online is fine 
during the day, so it's all good. Of course during certain days of the month 
daytime dev/test batch suffers for a few hours, but the MLC cost savings makes 
impacting those lower importance workloads worthwhile.

Scott Chapman

On Wed, 7 May 2014 18:37:03 -0500, Shane Ginnane <[email protected]> wrote:

>On Wed, 7 May 2014 10:47:49 -0400, John Eells wrote:
>
>>I'm not sure we have a white paper. 
>
>I would think a search of Techdocs for "Kathy Walsh" would be a good start.
>But of course we need to define whose "100%" we're talking about. z/OS has a 
>peculiar notion of CPU% - more so in virtualised environments like PR/SM 
>and/or z/VM.
>Then we have dynamic weighting issues - 100% busy of what proportion of the 
>machine ?. When ?. For how long ?.
>
>Nothing is as simple as it once was when we ran on bare metal.
>
>Shane ...
>
>----------------------------------------------------------------------
>For IBM-MAIN subscribe / signoff / archive access instructions,
>send email to [email protected] with the message: INFO IBM-MAIN

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to